Nearly every discovery call starts the same way: a business owner has been told, by a vendor or a well-meaning consultant, that they need "AI" in their product or their process. Sometimes that's true. More often, the honest answer is that a $30-a-month tool already does the job, and the custom build would be a slower, more expensive way to arrive at the same result.
Here's the framework we actually use — not to talk anyone out of a project, but to make sure the money goes where it pays off.
Start with the workflow, not the technology
The wrong question is "should we use AI for this?" The right question is "what does this workflow actually need?" Write down the steps a person currently takes, start to finish, including the annoying manual ones — the copy-paste, the re-typing, the double-checking. AI is one possible answer to a slow step. It is not the goal.
Three questions that decide it
1. Does an off-the-shelf tool already do 80% of this? If a mainstream SaaS product covers most of the workflow, buy it. Paying a subscription for something a team already built, tested, and maintains is almost always cheaper than reproducing it — even including the 20% it doesn't handle well. Custom work should target the last 20%, not the whole thing.
2. Is the workflow actually yours, or is it the industry standard? Payroll, invoicing, and basic scheduling are solved problems — every business does them roughly the same way, so the software written for "every business" fits fine. But if your process is genuinely shaped by something specific to your business — your data, your compliance obligations, your client relationships — a generic tool will always feel like it's fighting you. That's the signal a custom build is worth it.
3. Does the volume justify the build cost? A task that eats 20 minutes a month isn't worth automating no matter how annoying it is. A task that eats 10 hours a week, done by someone billed at $60/hour, pays for a modest build in a matter of months. Do the math before the meeting, not after.
Do the math before the meetingWhat "buy" actually looks like in practice
When the answer is buy, we say so — even in a paid discovery session, even when it means no engagement follows. Zapier, a vertical SaaS tool, or a $20/month AI writing assistant is frequently the right call for a small team, and pretending otherwise to sell a project is exactly the kind of vendor behavior that gives "AI consulting" a bad name.
What "build" actually looks like in practice
When the answer is build, it's usually because the workflow touches your specific data model, needs to talk to two or three systems that don't have a native integration, or handles something sensitive enough that you want full control over where the data goes and how it's logged. In those cases, a scoped, fixed-price build tends to cost less over 18 months than stacking subscriptions and workarounds to bend a generic tool into the same shape.
The test you can run yourself
Before you take a build proposal from anyone — us included — ask them to walk through this same logic with your specific workflow, out loud, before they quote you a number. If they skip straight to the price, that's worth noticing.